Database Virtualization: A New Frontier for Database Tuning and Physical Design
- 1 April 2007
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 388-394
Abstract
Resource virtualization is currently being employed at all levels of the IT infrastructure to improve provisioning and manageability, with the goal of reducing total cost of ownership. This means that database systems will increasingly be run in virtualized environments, inside virtual machines. This has many benefits, but it also introduces new tuning and physical design problems that are of interest to the database research community. In this paper, we discuss how virtualization can benefit database systems, and we present the tuning problems it introduces, which relate to setting the new "tuning knobs" that control resource allocation to virtual machines in the virtualized environment. We present a formulation of the visualization design problem, which focuses on setting resource allocation levels for different database workloads statically at deployment and configuration time. An important component of the solution to this problem is modeling the cost of a workload for a given resource allocation. We present an approach to this cost modeling that relies on using the query optimizer in a special virtualization-aware "what-if" mode. We also discuss the next steps in solving this problem, and present some long-term research directions.Keywords
This publication has 12 references indexed in Scilit:
- Continuous resource monitoring for self-predicting DBMS2005
- Virtual machine monitors: current technology and future trendsComputer, 2005
- The architecture of virtual machinesComputer, 2005
- Priority mechanisms for OLTP and transactional Web applications2004
- Automating layout of relational databases2004
- Xen and the art of virtualization2003
- Self-tuning Database Technology and Information Services2002
- Dynamic Reconfiguration Algorithm: Dynamically Tuning Multiple Buffer Pools2000
- Multi-dimensional resource scheduling for parallel queriesACM SIGMOD Record, 1996
- Dynamic resource brokering for multi-user query execution1995